web-research

# Web Research with x402 APIs Access Exa (neural search) and Firecrawl (web scraping) through x402-protected endpoints. ## Setup See [rules/getting-started.md](rules/getting-started.md) for installation and wallet setup. ## Quick Reference | Task | Endpoint | Price | Best For | |------|----------|-------|----------| | Neural search | `https://stableenrich.dev/api/exa/search` | $0.01 | Semantic web search | | Find similar | `https://stableenrich.dev/api/exa/find-similar` | $0.01 | Pages similar to a URL | | Extract text | `https://stableenrich.dev/api/exa/contents` | $0.002 | Clean text from URLs | | Direct answers | `https://stableenrich.dev/api/exa/answer` | $0.01 | Factual Q&A | | Scrape page | `https://stableenrich.dev/api/firecrawl/scrape` | $0.0126 | Single page to markdown | | Web search | `https://stableenrich.dev/api/firecrawl/search` | $0.0252 | Search with content snippets | | Crawl website | `https://stableenrich.dev/api/cloudflare/crawl` | $0.10 | Multi-page site crawl | | Poll crawl | `GET https://stableenrich.dev/api/cloudflare/jobs?token=...` | Free | Poll crawl results | ## When to Use What | Scenario | Tool | |----------|------| | General web search | WebSearch (free) or Exa ($0.01) | | Semantic/conceptual search | Exa search | | Find pages like X | Exa find-similar | | Get clean text from URL | Exa contents | | Scrape blocked/JS-heavy site | Firecrawl scrape | | Search + content snippets | Firecrawl search | | Quick fact lookup | Exa answer | | Crawl entire site/section | Cloudflare crawl | See [rules/when-to-use.md](rules/when-to-use.md) for detailed guidance. ## Exa Neural Search Semantic search that understands meaning, not just keywords: ```mcp agentcash.fetch( url="https://stableenrich.dev/api/exa/search", method="POST", body={ "query": "startups building AI agents for customer support", "numResults": 10 } ) ``` **Options:** - `query` - Search query (required) - `numResults` - Number of results (default: 5, max: 100) - `type` - Search mode/latency strategy ("auto", "fast", "deep"). Do not put content verticals here — use `category` - `includeDomains` - Only search these domains (full hostnames; some domains like reddit.com, x.com, nytimes.com are blocked and return 400) - `excludeDomains` - Skip these domains (same blocked-domain restriction applies) - `startPublishedDate` / `endPublishedDate` - Date range filter - `category` - Filter by content type: "company", "people", "research paper", "news", "pdf", "personal site", "financial report" (other strings are used as category hints) - Tip: Use `category: "people"` for people/profile discovery **Returns**: List of URLs with titles, snippets, and relevance scores. ## Find Similar Pages Find pages semantically similar to a reference URL: ```mcp agentcash.fetch( url="https://stableenrich.dev/api/exa/find-similar", method="POST", body={ "url": "https://example.com/article-i-like", "numResults": 10 } ) ``` Great for: - Finding competitor products - Discovering related content - Expanding research sources ## Extract Text Content Get clean, structured text from URLs: ```mcp agentcash.fetch( url="https://stableenrich.dev/api/exa/contents", method="POST", body={ "urls": [ "https://example.com/article1", "https://example.com/article2" ] } ) ``` **Options:** - `urls` - Array of URLs to extract - `text` - Include full text (default: true) - `highlights` - Include key highlights Cheapest option ($0.002) when you already have URLs and just need the content. ## Direct Answers Get factual answers to questions: ```mcp agentcash.fetch( url="https://stableenrich.dev/api/exa/answer", method="POST", body={ "query": "What is the population of Tokyo?" } ) ``` Returns a direct answer with source citations. Best for: - Factual questions - Quick lookups - Verification of claims ## Firecrawl Scrape Scrape a single page to clean markdown: ```mcp agentcash.fetch( url="https://stableenrich.dev/api/firecrawl/scrape", method="POST", body={ "url": "https://example.com/page-to-scrape" } ) ``` **Options:** - `url` - Page to scrape (required, the only parameter) **Returns**: `url` (final URL after redirects), `title`, and `content` (page as markdown). **Advantages over WebFetch:** - Handles JavaScript-rendered content - Bypasses common blocking - Extracts main content only - LLM-optimized markdown output ## Firecrawl Search Web search with automatic scraping of results: ```mcp agentcash.fetch( url="https://stableenrich.dev/api/firecrawl/search", method="POST", body={ "query": "best practices for react server components", "limit": 5 } ) ``` **Options:** - `query` - Search query (required) - `limit` - Number of results (default: 5, max: 10) Returns results with title, URL, description, and a content snippet (first ~500 chars of markdown) for each. ## Cloudflare Website Crawl Crawl multiple pages from a website with browser rendering. Async two-step pattern. **Step 1: Start the crawl (paid, $0.10)** ```mcp agentcash.fetch( url="https://stableenrich.dev/api/cloudflare/crawl", method="POST", body={ "url": "https://example.com", "limit": 10, "depth": 1, "formats": ["markdown"] } ) ``` Returns 202 with `{"token": "jwt..."}`. **Step 2: Poll for results (SIWX, free)** ```mcp agentcash.fetch( url="https://stableenrich.dev/api/cloudflare/jobs?token=JWT_TOKEN", method="GET" ) ``` Poll every 3-5 seconds until complete. **Parameters:** - `url` (required) — starting URL - `limit` (default 10, max 25) — max pages - `depth` (default 1, max 3) — max link depth - `formats` — `["markdown", "html", "json"]` - `render` (default false) — execute JavaScript - `source` — URL discovery source: "all", "sitemaps", or "links" - `options.includePatterns` / `excludePatterns` — URL wildcards (exclude wins); `options.includeExternalLinks` / `includeSubdomains` — scope flags Good for: crawling docs sites, scraping multiple pages, building sitemaps. ## Workflows ### Deep Research - [ ] (Optional) Check balance: `agentcash.get_balance` - [ ] Search broadly with Exa - [ ] Find related sources with find-similar - [ ] Extract content from top sources - [ ] Synthesize findings ```mcp agentcash.fetch( url="https://stableenrich.dev/api/exa/search", method="POST", body={"query": "AI agents in healthcare 2024", "numResults": 15} ) ``` ```mcp agentcash.fetch( url="https://stableenrich.dev/api/exa/find-similar", method="POST", body={"url": "https://best-article-found.com"} ) ``` ```mcp agentcash.fetch( url="https://stableenrich.dev/api/exa/contents", method="POST", body={"urls": ["url1", "url2", "url3"]} ) ``` ### Blocked Site Scraping - [ ] Try WebFetch first (free) - [ ] If blocked/empty, use Firecrawl (full JS rendering) for JS-heavy sites ```mcp agentcash.fetch( url="https://stableenrich.dev/api/firecrawl/scrape", method="POST", body={"url": "https://blocked-site.com/article"} ) ``` ## Cost Optimization - **Use Exa contents** ($0.002) when you already have URLs - **Use WebSearch/WebFetch first** (free) and fall back to x402 endpoints - **Batch URL extraction** - pass multiple URLs to Exa contents - **Limit results** - request only as many as needed ## Parallel Calls Independent searches can run in parallel: ```mcp # These don't depend on each other agentcash.fetch(url=".../exa/search", body={"query": "topic A"}) agentcash.fetch(url=".../exa/search", body={"query": "topic B"}) ```